[GFS2] Make journaled data files identical to normal files on disk
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / gfs2 / inode.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License v.2.
8 */
9
10#include <linux/sched.h>
11#include <linux/slab.h>
12#include <linux/spinlock.h>
13#include <linux/completion.h>
14#include <linux/buffer_head.h>
15#include <linux/posix_acl.h>
16#include <linux/sort.h>
17#include <asm/semaphore.h>
18
19#include "gfs2.h"
20#include "acl.h"
21#include "bmap.h"
22#include "dir.h"
23#include "eattr.h"
24#include "glock.h"
25#include "glops.h"
26#include "inode.h"
27#include "log.h"
28#include "meta_io.h"
29#include "ops_address.h"
30#include "ops_file.h"
31#include "ops_inode.h"
32#include "quota.h"
33#include "rgrp.h"
34#include "trans.h"
35#include "unlinked.h"
36
37/**
38 * inode_attr_in - Copy attributes from the dinode into the VFS inode
39 * @ip: The GFS2 inode (with embedded disk inode data)
40 * @inode: The Linux VFS inode
41 *
42 */
43
44static void inode_attr_in(struct gfs2_inode *ip, struct inode *inode)
45{
46 inode->i_ino = ip->i_num.no_formal_ino;
47
48 switch (ip->i_di.di_mode & S_IFMT) {
49 case S_IFBLK:
50 case S_IFCHR:
51 inode->i_rdev = MKDEV(ip->i_di.di_major, ip->i_di.di_minor);
52 break;
53 default:
54 inode->i_rdev = 0;
55 break;
56 };
57
58 inode->i_mode = ip->i_di.di_mode;
59 inode->i_nlink = ip->i_di.di_nlink;
60 inode->i_uid = ip->i_di.di_uid;
61 inode->i_gid = ip->i_di.di_gid;
62 i_size_write(inode, ip->i_di.di_size);
63 inode->i_atime.tv_sec = ip->i_di.di_atime;
64 inode->i_mtime.tv_sec = ip->i_di.di_mtime;
65 inode->i_ctime.tv_sec = ip->i_di.di_ctime;
66 inode->i_atime.tv_nsec = 0;
67 inode->i_mtime.tv_nsec = 0;
68 inode->i_ctime.tv_nsec = 0;
69 inode->i_blksize = PAGE_SIZE;
70 inode->i_blocks = ip->i_di.di_blocks <<
71 (ip->i_sbd->sd_sb.sb_bsize_shift - GFS2_BASIC_BLOCK_SHIFT);
72
73 if (ip->i_di.di_flags & GFS2_DIF_IMMUTABLE)
74 inode->i_flags |= S_IMMUTABLE;
75 else
76 inode->i_flags &= ~S_IMMUTABLE;
77
78 if (ip->i_di.di_flags & GFS2_DIF_APPENDONLY)
79 inode->i_flags |= S_APPEND;
80 else
81 inode->i_flags &= ~S_APPEND;
82}
83
84/**
85 * gfs2_inode_attr_in - Copy attributes from the dinode into the VFS inode
86 * @ip: The GFS2 inode (with embedded disk inode data)
87 *
88 */
89
90void gfs2_inode_attr_in(struct gfs2_inode *ip)
91{
92 struct inode *inode;
93
94 inode = gfs2_ip2v_lookup(ip);
95 if (inode) {
96 inode_attr_in(ip, inode);
97 iput(inode);
98 }
99}
100
101/**
102 * gfs2_inode_attr_out - Copy attributes from VFS inode into the dinode
103 * @ip: The GFS2 inode
104 *
105 * Only copy out the attributes that we want the VFS layer
106 * to be able to modify.
107 */
108
109void gfs2_inode_attr_out(struct gfs2_inode *ip)
110{
111 struct inode *inode = ip->i_vnode;
112
113 gfs2_assert_withdraw(ip->i_sbd,
114 (ip->i_di.di_mode & S_IFMT) == (inode->i_mode & S_IFMT));
115 ip->i_di.di_mode = inode->i_mode;
116 ip->i_di.di_uid = inode->i_uid;
117 ip->i_di.di_gid = inode->i_gid;
118 ip->i_di.di_atime = inode->i_atime.tv_sec;
119 ip->i_di.di_mtime = inode->i_mtime.tv_sec;
120 ip->i_di.di_ctime = inode->i_ctime.tv_sec;
121}
122
123/**
124 * gfs2_ip2v_lookup - Get the struct inode for a struct gfs2_inode
125 * @ip: the struct gfs2_inode to get the struct inode for
126 *
127 * Returns: A VFS inode, or NULL if none
128 */
129
130struct inode *gfs2_ip2v_lookup(struct gfs2_inode *ip)
131{
132 struct inode *inode = NULL;
133
134 gfs2_assert_warn(ip->i_sbd, test_bit(GIF_MIN_INIT, &ip->i_flags));
135
136 spin_lock(&ip->i_spin);
137 if (ip->i_vnode)
138 inode = igrab(ip->i_vnode);
139 spin_unlock(&ip->i_spin);
140
141 return inode;
142}
143
144/**
145 * gfs2_ip2v - Get/Create a struct inode for a struct gfs2_inode
146 * @ip: the struct gfs2_inode to get the struct inode for
147 *
148 * Returns: A VFS inode, or NULL if no mem
149 */
150
151struct inode *gfs2_ip2v(struct gfs2_inode *ip)
152{
153 struct inode *inode, *tmp;
154
155 inode = gfs2_ip2v_lookup(ip);
156 if (inode)
157 return inode;
158
159 tmp = new_inode(ip->i_sbd->sd_vfs);
160 if (!tmp)
161 return NULL;
162
163 inode_attr_in(ip, tmp);
164
165 if (S_ISREG(ip->i_di.di_mode)) {
166 tmp->i_op = &gfs2_file_iops;
167 tmp->i_fop = &gfs2_file_fops;
168 tmp->i_mapping->a_ops = &gfs2_file_aops;
169 } else if (S_ISDIR(ip->i_di.di_mode)) {
170 tmp->i_op = &gfs2_dir_iops;
171 tmp->i_fop = &gfs2_dir_fops;
172 } else if (S_ISLNK(ip->i_di.di_mode)) {
173 tmp->i_op = &gfs2_symlink_iops;
174 } else {
175 tmp->i_op = &gfs2_dev_iops;
176 init_special_inode(tmp, tmp->i_mode, tmp->i_rdev);
177 }
178
179 set_v2ip(tmp, NULL);
180
181 for (;;) {
182 spin_lock(&ip->i_spin);
183 if (!ip->i_vnode)
184 break;
185 inode = igrab(ip->i_vnode);
186 spin_unlock(&ip->i_spin);
187
188 if (inode) {
189 iput(tmp);
190 return inode;
191 }
192 yield();
193 }
194
195 inode = tmp;
196
197 gfs2_inode_hold(ip);
198 ip->i_vnode = inode;
199 set_v2ip(inode, ip);
200
201 spin_unlock(&ip->i_spin);
202
203 insert_inode_hash(inode);
204
205 return inode;
206}
207
208static int iget_test(struct inode *inode, void *opaque)
209{
210 struct gfs2_inode *ip = get_v2ip(inode);
211 struct gfs2_inum *inum = (struct gfs2_inum *)opaque;
212
213 if (ip && ip->i_num.no_addr == inum->no_addr)
214 return 1;
215
216 return 0;
217}
218
219struct inode *gfs2_iget(struct super_block *sb, struct gfs2_inum *inum)
220{
221 return ilookup5(sb, (unsigned long)inum->no_formal_ino,
222 iget_test, inum);
223}
224
225void gfs2_inode_min_init(struct gfs2_inode *ip, unsigned int type)
226{
227 spin_lock(&ip->i_spin);
228 if (!test_and_set_bit(GIF_MIN_INIT, &ip->i_flags)) {
229 ip->i_di.di_nlink = 1;
230 ip->i_di.di_mode = DT2IF(type);
231 }
232 spin_unlock(&ip->i_spin);
233}
234
235/**
236 * gfs2_inode_refresh - Refresh the incore copy of the dinode
237 * @ip: The GFS2 inode
238 *
239 * Returns: errno
240 */
241
242int gfs2_inode_refresh(struct gfs2_inode *ip)
243{
244 struct buffer_head *dibh;
245 int error;
246
247 error = gfs2_meta_inode_buffer(ip, &dibh);
248 if (error)
249 return error;
250
251 if (gfs2_metatype_check(ip->i_sbd, dibh, GFS2_METATYPE_DI)) {
252 brelse(dibh);
253 return -EIO;
254 }
255
256 spin_lock(&ip->i_spin);
257 gfs2_dinode_in(&ip->i_di, dibh->b_data);
258 set_bit(GIF_MIN_INIT, &ip->i_flags);
259 spin_unlock(&ip->i_spin);
260
261 brelse(dibh);
262
263 if (ip->i_num.no_addr != ip->i_di.di_num.no_addr) {
264 if (gfs2_consist_inode(ip))
265 gfs2_dinode_print(&ip->i_di);
266 return -EIO;
267 }
268 if (ip->i_num.no_formal_ino != ip->i_di.di_num.no_formal_ino)
269 return -ESTALE;
270
271 ip->i_vn = ip->i_gl->gl_vn;
272
273 return 0;
274}
275
276/**
277 * inode_create - create a struct gfs2_inode
278 * @i_gl: The glock covering the inode
279 * @inum: The inode number
280 * @io_gl: the iopen glock to acquire/hold (using holder in new gfs2_inode)
281 * @io_state: the state the iopen glock should be acquired in
282 * @ipp: pointer to put the returned inode in
283 *
284 * Returns: errno
285 */
286
287static int inode_create(struct gfs2_glock *i_gl, struct gfs2_inum *inum,
288 struct gfs2_glock *io_gl, unsigned int io_state,
289 struct gfs2_inode **ipp)
290{
291 struct gfs2_sbd *sdp = i_gl->gl_sbd;
292 struct gfs2_inode *ip;
293 int error = 0;
294
295 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL);
296 if (!ip)
297 return -ENOMEM;
298 memset(ip, 0, sizeof(struct gfs2_inode));
299
300 ip->i_num = *inum;
301
302 atomic_set(&ip->i_count, 1);
303
304 ip->i_vn = i_gl->gl_vn - 1;
305
306 ip->i_gl = i_gl;
307 ip->i_sbd = sdp;
308
309 spin_lock_init(&ip->i_spin);
310 init_rwsem(&ip->i_rw_mutex);
311
312 ip->i_greedy = gfs2_tune_get(sdp, gt_greedy_default);
313
314 error = gfs2_glock_nq_init(io_gl,
315 io_state, GL_LOCAL_EXCL | GL_EXACT,
316 &ip->i_iopen_gh);
317 if (error)
318 goto fail;
319 ip->i_iopen_gh.gh_owner = NULL;
320
321 spin_lock(&io_gl->gl_spin);
322 gfs2_glock_hold(i_gl);
323 set_gl2gl(io_gl, i_gl);
324 spin_unlock(&io_gl->gl_spin);
325
326 gfs2_glock_hold(i_gl);
327 set_gl2ip(i_gl, ip);
328
329 atomic_inc(&sdp->sd_inode_count);
330
331 *ipp = ip;
332
333 return 0;
334
335 fail:
336 gfs2_meta_cache_flush(ip);
337 kmem_cache_free(gfs2_inode_cachep, ip);
338 *ipp = NULL;
339
340 return error;
341}
342
343/**
344 * gfs2_inode_get - Create or get a reference on an inode
345 * @i_gl: The glock covering the inode
346 * @inum: The inode number
347 * @create:
348 * @ipp: pointer to put the returned inode in
349 *
350 * Returns: errno
351 */
352
353int gfs2_inode_get(struct gfs2_glock *i_gl, struct gfs2_inum *inum, int create,
354 struct gfs2_inode **ipp)
355{
356 struct gfs2_sbd *sdp = i_gl->gl_sbd;
357 struct gfs2_glock *io_gl;
358 int error = 0;
359
360 gfs2_glmutex_lock(i_gl);
361
362 *ipp = get_gl2ip(i_gl);
363 if (*ipp) {
364 error = -ESTALE;
365 if ((*ipp)->i_num.no_formal_ino != inum->no_formal_ino)
366 goto out;
367 atomic_inc(&(*ipp)->i_count);
368 error = 0;
369 goto out;
370 }
371
372 if (!create)
373 goto out;
374
375 error = gfs2_glock_get(sdp, inum->no_addr, &gfs2_iopen_glops,
376 CREATE, &io_gl);
377 if (!error) {
378 error = inode_create(i_gl, inum, io_gl, LM_ST_SHARED, ipp);
379 gfs2_glock_put(io_gl);
380 }
381
382 out:
383 gfs2_glmutex_unlock(i_gl);
384
385 return error;
386}
387
388void gfs2_inode_hold(struct gfs2_inode *ip)
389{
390 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
391 atomic_inc(&ip->i_count);
392}
393
394void gfs2_inode_put(struct gfs2_inode *ip)
395{
396 gfs2_assert(ip->i_sbd, atomic_read(&ip->i_count) > 0);
397 atomic_dec(&ip->i_count);
398}
399
400void gfs2_inode_destroy(struct gfs2_inode *ip)
401{
402 struct gfs2_sbd *sdp = ip->i_sbd;
403 struct gfs2_glock *io_gl = ip->i_iopen_gh.gh_gl;
404 struct gfs2_glock *i_gl = ip->i_gl;
405
406 gfs2_assert_warn(sdp, !atomic_read(&ip->i_count));
407 gfs2_assert(sdp, get_gl2gl(io_gl) == i_gl);
408
409 spin_lock(&io_gl->gl_spin);
410 set_gl2gl(io_gl, NULL);
411 gfs2_glock_put(i_gl);
412 spin_unlock(&io_gl->gl_spin);
413
414 gfs2_glock_dq_uninit(&ip->i_iopen_gh);
415
416 gfs2_meta_cache_flush(ip);
417 kmem_cache_free(gfs2_inode_cachep, ip);
418
419 set_gl2ip(i_gl, NULL);
420 gfs2_glock_put(i_gl);
421
422 atomic_dec(&sdp->sd_inode_count);
423}
424
425static int dinode_dealloc(struct gfs2_inode *ip, struct gfs2_unlinked *ul)
426{
427 struct gfs2_sbd *sdp = ip->i_sbd;
428 struct gfs2_alloc *al;
429 struct gfs2_rgrpd *rgd;
430 int error;
431
432 if (ip->i_di.di_blocks != 1) {
433 if (gfs2_consist_inode(ip))
434 gfs2_dinode_print(&ip->i_di);
435 return -EIO;
436 }
437
438 al = gfs2_alloc_get(ip);
439
440 error = gfs2_quota_hold(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
441 if (error)
442 goto out;
443
444 error = gfs2_rindex_hold(sdp, &al->al_ri_gh);
445 if (error)
446 goto out_qs;
447
448 rgd = gfs2_blk2rgrpd(sdp, ip->i_num.no_addr);
449 if (!rgd) {
450 gfs2_consist_inode(ip);
451 error = -EIO;
452 goto out_rindex_relse;
453 }
454
455 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0,
456 &al->al_rgd_gh);
457 if (error)
458 goto out_rindex_relse;
459
460 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
461 RES_STATFS + RES_QUOTA, 1);
462 if (error)
463 goto out_rg_gunlock;
464
465 gfs2_trans_add_gl(ip->i_gl);
466
467 gfs2_free_di(rgd, ip);
468
469 error = gfs2_unlinked_ondisk_rm(sdp, ul);
470
471 gfs2_trans_end(sdp);
472 clear_bit(GLF_STICKY, &ip->i_gl->gl_flags);
473
474 out_rg_gunlock:
475 gfs2_glock_dq_uninit(&al->al_rgd_gh);
476
477 out_rindex_relse:
478 gfs2_glock_dq_uninit(&al->al_ri_gh);
479
480 out_qs:
481 gfs2_quota_unhold(ip);
482
483 out:
484 gfs2_alloc_put(ip);
485
486 return error;
487}
488
489/**
490 * inode_dealloc - Deallocate all on-disk blocks for an inode (dinode)
491 * @sdp: the filesystem
492 * @inum: the inode number to deallocate
493 * @io_gh: a holder for the iopen glock for this inode
494 *
495 * Returns: errno
496 */
497
498static int inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul,
499 struct gfs2_holder *io_gh)
500{
501 struct gfs2_inode *ip;
502 struct gfs2_holder i_gh;
503 int error;
504
505 error = gfs2_glock_nq_num(sdp,
506 ul->ul_ut.ut_inum.no_addr, &gfs2_inode_glops,
507 LM_ST_EXCLUSIVE, 0, &i_gh);
508 if (error)
509 return error;
510
511 /* We reacquire the iopen lock here to avoid a race with the NFS server
512 calling gfs2_read_inode() with the inode number of a inode we're in
513 the process of deallocating. And we can't keep our hold on the lock
514 from inode_dealloc_init() for deadlock reasons. */
515
516 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY, io_gh);
517 error = gfs2_glock_nq(io_gh);
518 switch (error) {
519 case 0:
520 break;
521 case GLR_TRYFAILED:
522 error = 1;
523 default:
524 goto out;
525 }
526
527 gfs2_assert_warn(sdp, !get_gl2ip(i_gh.gh_gl));
528 error = inode_create(i_gh.gh_gl, &ul->ul_ut.ut_inum, io_gh->gh_gl,
529 LM_ST_EXCLUSIVE, &ip);
530
531 gfs2_glock_dq(io_gh);
532
533 if (error)
534 goto out;
535
536 error = gfs2_inode_refresh(ip);
537 if (error)
538 goto out_iput;
539
540 if (ip->i_di.di_nlink) {
541 if (gfs2_consist_inode(ip))
542 gfs2_dinode_print(&ip->i_di);
543 error = -EIO;
544 goto out_iput;
545 }
546
547 if (S_ISDIR(ip->i_di.di_mode) &&
548 (ip->i_di.di_flags & GFS2_DIF_EXHASH)) {
549 error = gfs2_dir_exhash_dealloc(ip);
550 if (error)
551 goto out_iput;
552 }
553
554 if (ip->i_di.di_eattr) {
555 error = gfs2_ea_dealloc(ip);
556 if (error)
557 goto out_iput;
558 }
559
560 if (!gfs2_is_stuffed(ip)) {
561 error = gfs2_file_dealloc(ip);
562 if (error)
563 goto out_iput;
564 }
565
566 error = dinode_dealloc(ip, ul);
567 if (error)
568 goto out_iput;
569
570 out_iput:
571 gfs2_glmutex_lock(i_gh.gh_gl);
572 gfs2_inode_put(ip);
573 gfs2_inode_destroy(ip);
574 gfs2_glmutex_unlock(i_gh.gh_gl);
575
576 out:
577 gfs2_glock_dq_uninit(&i_gh);
578
579 return error;
580}
581
582/**
583 * try_inode_dealloc - Try to deallocate an inode and all its blocks
584 * @sdp: the filesystem
585 *
586 * Returns: 0 on success, -errno on error, 1 on busy (inode open)
587 */
588
589static int try_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
590{
591 struct gfs2_holder io_gh;
592 int error = 0;
593
594 gfs2_try_toss_inode(sdp, &ul->ul_ut.ut_inum);
595
596 error = gfs2_glock_nq_num(sdp,
597 ul->ul_ut.ut_inum.no_addr, &gfs2_iopen_glops,
598 LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB, &io_gh);
599 switch (error) {
600 case 0:
601 break;
602 case GLR_TRYFAILED:
603 return 1;
604 default:
605 return error;
606 }
607
608 gfs2_glock_dq(&io_gh);
609 error = inode_dealloc(sdp, ul, &io_gh);
610 gfs2_holder_uninit(&io_gh);
611
612 return error;
613}
614
615static int inode_dealloc_uninit(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
616{
617 struct gfs2_rgrpd *rgd;
618 struct gfs2_holder ri_gh, rgd_gh;
619 int error;
620
621 error = gfs2_rindex_hold(sdp, &ri_gh);
622 if (error)
623 return error;
624
625 rgd = gfs2_blk2rgrpd(sdp, ul->ul_ut.ut_inum.no_addr);
626 if (!rgd) {
627 gfs2_consist(sdp);
628 error = -EIO;
629 goto out;
630 }
631
632 error = gfs2_glock_nq_init(rgd->rd_gl, LM_ST_EXCLUSIVE, 0, &rgd_gh);
633 if (error)
634 goto out;
635
636 error = gfs2_trans_begin(sdp,
637 RES_RG_BIT + RES_UNLINKED + RES_STATFS,
638 0);
639 if (error)
640 goto out_gunlock;
641
642 gfs2_free_uninit_di(rgd, ul->ul_ut.ut_inum.no_addr);
643 gfs2_unlinked_ondisk_rm(sdp, ul);
644
645 gfs2_trans_end(sdp);
646
647 out_gunlock:
648 gfs2_glock_dq_uninit(&rgd_gh);
649 out:
650 gfs2_glock_dq_uninit(&ri_gh);
651
652 return error;
653}
654
655int gfs2_inode_dealloc(struct gfs2_sbd *sdp, struct gfs2_unlinked *ul)
656{
657 if (ul->ul_ut.ut_flags & GFS2_UTF_UNINIT)
658 return inode_dealloc_uninit(sdp, ul);
659 else
660 return try_inode_dealloc(sdp, ul);
661}
662
663/**
664 * gfs2_change_nlink - Change nlink count on inode
665 * @ip: The GFS2 inode
666 * @diff: The change in the nlink count required
667 *
668 * Returns: errno
669 */
670
671int gfs2_change_nlink(struct gfs2_inode *ip, int diff)
672{
673 struct buffer_head *dibh;
674 uint32_t nlink;
675 int error;
676
677 nlink = ip->i_di.di_nlink + diff;
678
679 /* If we are reducing the nlink count, but the new value ends up being
680 bigger than the old one, we must have underflowed. */
681 if (diff < 0 && nlink > ip->i_di.di_nlink) {
682 if (gfs2_consist_inode(ip))
683 gfs2_dinode_print(&ip->i_di);
684 return -EIO;
685 }
686
687 error = gfs2_meta_inode_buffer(ip, &dibh);
688 if (error)
689 return error;
690
691 ip->i_di.di_nlink = nlink;
692 ip->i_di.di_ctime = get_seconds();
693
d4e9c4c3 694 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
b3b94faa
DT
695 gfs2_dinode_out(&ip->i_di, dibh->b_data);
696 brelse(dibh);
697
698 return 0;
699}
700
701/**
702 * gfs2_lookupi - Look up a filename in a directory and return its inode
703 * @d_gh: An initialized holder for the directory glock
704 * @name: The name of the inode to look for
705 * @is_root: If 1, ignore the caller's permissions
706 * @i_gh: An uninitialized holder for the new inode glock
707 *
708 * There will always be a vnode (Linux VFS inode) for the d_gh inode unless
709 * @is_root is true.
710 *
711 * Returns: errno
712 */
713
714int gfs2_lookupi(struct gfs2_inode *dip, struct qstr *name, int is_root,
715 struct gfs2_inode **ipp)
716{
717 struct gfs2_sbd *sdp = dip->i_sbd;
718 struct gfs2_holder d_gh;
719 struct gfs2_inum inum;
720 unsigned int type;
721 struct gfs2_glock *gl;
722 int error;
723
724 if (!name->len || name->len > GFS2_FNAMESIZE)
725 return -ENAMETOOLONG;
726
727 if (gfs2_filecmp(name, ".", 1) ||
f42faf4f 728 (gfs2_filecmp(name, "..", 2) && dip == get_v2ip(sdp->sd_root_dir))) {
b3b94faa
DT
729 gfs2_inode_hold(dip);
730 *ipp = dip;
731 return 0;
732 }
733
734 error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh);
735 if (error)
736 return error;
737
738 if (!is_root) {
739 error = gfs2_repermission(dip->i_vnode, MAY_EXEC, NULL);
740 if (error)
741 goto out;
742 }
743
744 error = gfs2_dir_search(dip, name, &inum, &type);
745 if (error)
746 goto out;
747
748 error = gfs2_glock_get(sdp, inum.no_addr, &gfs2_inode_glops,
749 CREATE, &gl);
750 if (error)
751 goto out;
752
753 error = gfs2_inode_get(gl, &inum, CREATE, ipp);
754 if (!error)
755 gfs2_inode_min_init(*ipp, type);
756
757 gfs2_glock_put(gl);
758
759 out:
760 gfs2_glock_dq_uninit(&d_gh);
761
762 return error;
763}
764
765static int pick_formal_ino_1(struct gfs2_sbd *sdp, uint64_t *formal_ino)
766{
f42faf4f 767 struct gfs2_inode *ip = get_v2ip(sdp->sd_ir_inode);
b3b94faa
DT
768 struct buffer_head *bh;
769 struct gfs2_inum_range ir;
770 int error;
771
772 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
773 if (error)
774 return error;
775 down(&sdp->sd_inum_mutex);
776
777 error = gfs2_meta_inode_buffer(ip, &bh);
778 if (error) {
779 up(&sdp->sd_inum_mutex);
780 gfs2_trans_end(sdp);
781 return error;
782 }
783
784 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
785
786 if (ir.ir_length) {
787 *formal_ino = ir.ir_start++;
788 ir.ir_length--;
d4e9c4c3 789 gfs2_trans_add_bh(ip->i_gl, bh, 1);
b3b94faa
DT
790 gfs2_inum_range_out(&ir,
791 bh->b_data + sizeof(struct gfs2_dinode));
792 brelse(bh);
793 up(&sdp->sd_inum_mutex);
794 gfs2_trans_end(sdp);
795 return 0;
796 }
797
798 brelse(bh);
799
800 up(&sdp->sd_inum_mutex);
801 gfs2_trans_end(sdp);
802
803 return 1;
804}
805
806static int pick_formal_ino_2(struct gfs2_sbd *sdp, uint64_t *formal_ino)
807{
f42faf4f
SW
808 struct gfs2_inode *ip = get_v2ip(sdp->sd_ir_inode);
809 struct gfs2_inode *m_ip = get_v2ip(sdp->sd_inum_inode);
b3b94faa
DT
810 struct gfs2_holder gh;
811 struct buffer_head *bh;
812 struct gfs2_inum_range ir;
813 int error;
814
815 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh);
816 if (error)
817 return error;
818
819 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0);
820 if (error)
821 goto out;
822 down(&sdp->sd_inum_mutex);
823
824 error = gfs2_meta_inode_buffer(ip, &bh);
825 if (error)
826 goto out_end_trans;
827
828 gfs2_inum_range_in(&ir, bh->b_data + sizeof(struct gfs2_dinode));
829
830 if (!ir.ir_length) {
831 struct buffer_head *m_bh;
832 uint64_t x, y;
833
834 error = gfs2_meta_inode_buffer(m_ip, &m_bh);
835 if (error)
836 goto out_brelse;
837
838 x = *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode));
839 x = y = be64_to_cpu(x);
840 ir.ir_start = x;
841 ir.ir_length = GFS2_INUM_QUANTUM;
842 x += GFS2_INUM_QUANTUM;
843 if (x < y)
844 gfs2_consist_inode(m_ip);
845 x = cpu_to_be64(x);
d4e9c4c3 846 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1);
b3b94faa
DT
847 *(uint64_t *)(m_bh->b_data + sizeof(struct gfs2_dinode)) = x;
848
849 brelse(m_bh);
850 }
851
852 *formal_ino = ir.ir_start++;
853 ir.ir_length--;
854
d4e9c4c3 855 gfs2_trans_add_bh(ip->i_gl, bh, 1);
b3b94faa
DT
856 gfs2_inum_range_out(&ir, bh->b_data + sizeof(struct gfs2_dinode));
857
858 out_brelse:
859 brelse(bh);
860
861 out_end_trans:
862 up(&sdp->sd_inum_mutex);
863 gfs2_trans_end(sdp);
864
865 out:
866 gfs2_glock_dq_uninit(&gh);
867
868 return error;
869}
870
871static int pick_formal_ino(struct gfs2_sbd *sdp, uint64_t *inum)
872{
873 int error;
874
875 error = pick_formal_ino_1(sdp, inum);
876 if (error <= 0)
877 return error;
878
879 error = pick_formal_ino_2(sdp, inum);
880
881 return error;
882}
883
884/**
885 * create_ok - OK to create a new on-disk inode here?
886 * @dip: Directory in which dinode is to be created
887 * @name: Name of new dinode
888 * @mode:
889 *
890 * Returns: errno
891 */
892
893static int create_ok(struct gfs2_inode *dip, struct qstr *name,
894 unsigned int mode)
895{
896 int error;
897
898 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
899 if (error)
900 return error;
901
902 /* Don't create entries in an unlinked directory */
903 if (!dip->i_di.di_nlink)
904 return -EPERM;
905
906 error = gfs2_dir_search(dip, name, NULL, NULL);
907 switch (error) {
908 case -ENOENT:
909 error = 0;
910 break;
911 case 0:
912 return -EEXIST;
913 default:
914 return error;
915 }
916
917 if (dip->i_di.di_entries == (uint32_t)-1)
918 return -EFBIG;
919 if (S_ISDIR(mode) && dip->i_di.di_nlink == (uint32_t)-1)
920 return -EMLINK;
921
922 return 0;
923}
924
925static void munge_mode_uid_gid(struct gfs2_inode *dip, unsigned int *mode,
926 unsigned int *uid, unsigned int *gid)
927{
928 if (dip->i_sbd->sd_args.ar_suiddir &&
929 (dip->i_di.di_mode & S_ISUID) &&
930 dip->i_di.di_uid) {
931 if (S_ISDIR(*mode))
932 *mode |= S_ISUID;
933 else if (dip->i_di.di_uid != current->fsuid)
934 *mode &= ~07111;
935 *uid = dip->i_di.di_uid;
936 } else
937 *uid = current->fsuid;
938
939 if (dip->i_di.di_mode & S_ISGID) {
940 if (S_ISDIR(*mode))
941 *mode |= S_ISGID;
942 *gid = dip->i_di.di_gid;
943 } else
944 *gid = current->fsgid;
945}
946
947static int alloc_dinode(struct gfs2_inode *dip, struct gfs2_unlinked *ul)
948{
949 struct gfs2_sbd *sdp = dip->i_sbd;
950 int error;
951
952 gfs2_alloc_get(dip);
953
954 dip->i_alloc.al_requested = RES_DINODE;
955 error = gfs2_inplace_reserve(dip);
956 if (error)
957 goto out;
958
959 error = gfs2_trans_begin(sdp, RES_RG_BIT + RES_UNLINKED +
960 RES_STATFS, 0);
961 if (error)
962 goto out_ipreserv;
963
964 ul->ul_ut.ut_inum.no_addr = gfs2_alloc_di(dip);
965
966 ul->ul_ut.ut_flags = GFS2_UTF_UNINIT;
967 error = gfs2_unlinked_ondisk_add(sdp, ul);
968
969 gfs2_trans_end(sdp);
970
971 out_ipreserv:
972 gfs2_inplace_release(dip);
973
974 out:
975 gfs2_alloc_put(dip);
976
977 return error;
978}
979
980/**
981 * init_dinode - Fill in a new dinode structure
982 * @dip: the directory this inode is being created in
983 * @gl: The glock covering the new inode
984 * @inum: the inode number
985 * @mode: the file permissions
986 * @uid:
987 * @gid:
988 *
989 */
990
991static void init_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
992 struct gfs2_inum *inum, unsigned int mode,
993 unsigned int uid, unsigned int gid)
994{
995 struct gfs2_sbd *sdp = dip->i_sbd;
b96ca4fa 996 struct gfs2_dinode *di;
b3b94faa
DT
997 struct buffer_head *dibh;
998
999 dibh = gfs2_meta_new(gl, inum->no_addr);
d4e9c4c3 1000 gfs2_trans_add_bh(gl, dibh, 1);
b3b94faa
DT
1001 gfs2_metatype_set(dibh, GFS2_METATYPE_DI, GFS2_FORMAT_DI);
1002 gfs2_buffer_clear_tail(dibh, sizeof(struct gfs2_dinode));
b96ca4fa
SW
1003 di = (struct gfs2_dinode *)dibh->b_data;
1004
2442a098
SW
1005 di->di_num.no_formal_ino = cpu_to_be64(inum->no_formal_ino);
1006 di->di_num.no_addr = cpu_to_be64(inum->no_addr);
b96ca4fa
SW
1007 di->di_mode = cpu_to_be32(mode);
1008 di->di_uid = cpu_to_be32(uid);
1009 di->di_gid = cpu_to_be32(gid);
1010 di->di_nlink = cpu_to_be32(0);
1011 di->di_size = cpu_to_be64(0);
1012 di->di_blocks = cpu_to_be64(1);
1013 di->di_atime = di->di_mtime = di->di_ctime = cpu_to_be64(get_seconds());
1014 di->di_major = di->di_minor = cpu_to_be32(0);
1015 di->di_goal_meta = di->di_goal_data = cpu_to_be64(inum->no_addr);
1016 di->__pad[0] = di->__pad[1] = 0;
1017 di->di_flags = cpu_to_be32(0);
b3b94faa
DT
1018
1019 if (S_ISREG(mode)) {
1020 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA) ||
1021 gfs2_tune_get(sdp, gt_new_files_jdata))
b96ca4fa 1022 di->di_flags |= cpu_to_be32(GFS2_DIF_JDATA);
b3b94faa
DT
1023 if ((dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO) ||
1024 gfs2_tune_get(sdp, gt_new_files_directio))
b96ca4fa 1025 di->di_flags |= cpu_to_be32(GFS2_DIF_DIRECTIO);
b3b94faa 1026 } else if (S_ISDIR(mode)) {
b96ca4fa
SW
1027 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & GFS2_DIF_INHERIT_DIRECTIO);
1028 di->di_flags |= cpu_to_be32(dip->i_di.di_flags & GFS2_DIF_INHERIT_JDATA);
b3b94faa
DT
1029 }
1030
b96ca4fa
SW
1031 di->__pad1 = 0;
1032 di->di_height = cpu_to_be32(0);
1033 di->__pad2 = 0;
1034 di->__pad3 = 0;
1035 di->di_depth = cpu_to_be16(0);
1036 di->di_entries = cpu_to_be32(0);
1037 memset(&di->__pad4, 0, sizeof(di->__pad4));
1038 di->di_eattr = cpu_to_be64(0);
1039 memset(&di->di_reserved, 0, sizeof(di->di_reserved));
1040
b3b94faa
DT
1041 brelse(dibh);
1042}
1043
1044static int make_dinode(struct gfs2_inode *dip, struct gfs2_glock *gl,
1045 unsigned int mode, struct gfs2_unlinked *ul)
1046{
1047 struct gfs2_sbd *sdp = dip->i_sbd;
1048 unsigned int uid, gid;
1049 int error;
1050
1051 munge_mode_uid_gid(dip, &mode, &uid, &gid);
1052
1053 gfs2_alloc_get(dip);
1054
1055 error = gfs2_quota_lock(dip, uid, gid);
1056 if (error)
1057 goto out;
1058
1059 error = gfs2_quota_check(dip, uid, gid);
1060 if (error)
1061 goto out_quota;
1062
1063 error = gfs2_trans_begin(sdp, RES_DINODE + RES_UNLINKED +
1064 RES_QUOTA, 0);
1065 if (error)
1066 goto out_quota;
1067
1068 ul->ul_ut.ut_flags = 0;
1069 error = gfs2_unlinked_ondisk_munge(sdp, ul);
1070
1071 init_dinode(dip, gl, &ul->ul_ut.ut_inum,
1072 mode, uid, gid);
1073
1074 gfs2_quota_change(dip, +1, uid, gid);
1075
1076 gfs2_trans_end(sdp);
1077
1078 out_quota:
1079 gfs2_quota_unlock(dip);
1080
1081 out:
1082 gfs2_alloc_put(dip);
1083
1084 return error;
1085}
1086
1087static int link_dinode(struct gfs2_inode *dip, struct qstr *name,
1088 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1089{
1090 struct gfs2_sbd *sdp = dip->i_sbd;
1091 struct gfs2_alloc *al;
1092 int alloc_required;
1093 struct buffer_head *dibh;
1094 int error;
1095
1096 al = gfs2_alloc_get(dip);
1097
1098 error = gfs2_quota_lock(dip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
1099 if (error)
1100 goto fail;
1101
1102 error = gfs2_diradd_alloc_required(dip, name, &alloc_required);
1103 if (alloc_required) {
1104 error = gfs2_quota_check(dip, dip->i_di.di_uid,
1105 dip->i_di.di_gid);
1106 if (error)
1107 goto fail_quota_locks;
1108
1109 al->al_requested = sdp->sd_max_dirres;
1110
1111 error = gfs2_inplace_reserve(dip);
1112 if (error)
1113 goto fail_quota_locks;
1114
1115 error = gfs2_trans_begin(sdp,
1116 sdp->sd_max_dirres +
1117 al->al_rgd->rd_ri.ri_length +
1118 2 * RES_DINODE + RES_UNLINKED +
1119 RES_STATFS + RES_QUOTA, 0);
1120 if (error)
1121 goto fail_ipreserv;
1122 } else {
1123 error = gfs2_trans_begin(sdp,
1124 RES_LEAF +
1125 2 * RES_DINODE +
1126 RES_UNLINKED, 0);
1127 if (error)
1128 goto fail_quota_locks;
1129 }
1130
1131 error = gfs2_dir_add(dip, name, &ip->i_num, IF2DT(ip->i_di.di_mode));
1132 if (error)
1133 goto fail_end_trans;
1134
1135 error = gfs2_meta_inode_buffer(ip, &dibh);
1136 if (error)
1137 goto fail_end_trans;
1138 ip->i_di.di_nlink = 1;
d4e9c4c3 1139 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
b3b94faa
DT
1140 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1141 brelse(dibh);
1142
1143 error = gfs2_unlinked_ondisk_rm(sdp, ul);
1144 if (error)
1145 goto fail_end_trans;
1146
1147 return 0;
1148
1149 fail_end_trans:
1150 gfs2_trans_end(sdp);
1151
1152 fail_ipreserv:
1153 if (dip->i_alloc.al_rgd)
1154 gfs2_inplace_release(dip);
1155
1156 fail_quota_locks:
1157 gfs2_quota_unlock(dip);
1158
1159 fail:
1160 gfs2_alloc_put(dip);
1161
1162 return error;
1163}
1164
1165/**
1166 * gfs2_createi - Create a new inode
1167 * @ghs: An array of two holders
1168 * @name: The name of the new file
1169 * @mode: the permissions on the new inode
1170 *
1171 * @ghs[0] is an initialized holder for the directory
1172 * @ghs[1] is the holder for the inode lock
1173 *
1174 * If the return value is 0, the glocks on both the directory and the new
1175 * file are held. A transaction has been started and an inplace reservation
1176 * is held, as well.
1177 *
1178 * Returns: errno
1179 */
1180
1181int gfs2_createi(struct gfs2_holder *ghs, struct qstr *name, unsigned int mode)
1182{
1183 struct gfs2_inode *dip = get_gl2ip(ghs->gh_gl);
1184 struct gfs2_sbd *sdp = dip->i_sbd;
1185 struct gfs2_unlinked *ul;
1186 struct gfs2_inode *ip;
1187 int error;
1188
1189 if (!name->len || name->len > GFS2_FNAMESIZE)
1190 return -ENAMETOOLONG;
1191
1192 error = gfs2_unlinked_get(sdp, &ul);
1193 if (error)
1194 return error;
1195
1196 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1197 error = gfs2_glock_nq(ghs);
1198 if (error)
1199 goto fail;
1200
1201 error = create_ok(dip, name, mode);
1202 if (error)
1203 goto fail_gunlock;
1204
1205 error = pick_formal_ino(sdp, &ul->ul_ut.ut_inum.no_formal_ino);
1206 if (error)
1207 goto fail_gunlock;
1208
1209 error = alloc_dinode(dip, ul);
1210 if (error)
1211 goto fail_gunlock;
1212
1213 if (ul->ul_ut.ut_inum.no_addr < dip->i_num.no_addr) {
1214 gfs2_glock_dq(ghs);
1215
1216 error = gfs2_glock_nq_num(sdp,
1217 ul->ul_ut.ut_inum.no_addr,
1218 &gfs2_inode_glops,
1219 LM_ST_EXCLUSIVE, GL_SKIP,
1220 ghs + 1);
1221 if (error) {
1222 gfs2_unlinked_put(sdp, ul);
1223 return error;
1224 }
1225
1226 gfs2_holder_reinit(LM_ST_EXCLUSIVE, 0, ghs);
1227 error = gfs2_glock_nq(ghs);
1228 if (error) {
1229 gfs2_glock_dq_uninit(ghs + 1);
1230 gfs2_unlinked_put(sdp, ul);
1231 return error;
1232 }
1233
1234 error = create_ok(dip, name, mode);
1235 if (error)
1236 goto fail_gunlock2;
1237 } else {
1238 error = gfs2_glock_nq_num(sdp,
1239 ul->ul_ut.ut_inum.no_addr,
1240 &gfs2_inode_glops,
1241 LM_ST_EXCLUSIVE, GL_SKIP,
1242 ghs + 1);
1243 if (error)
1244 goto fail_gunlock;
1245 }
1246
1247 error = make_dinode(dip, ghs[1].gh_gl, mode, ul);
1248 if (error)
1249 goto fail_gunlock2;
1250
1251 error = gfs2_inode_get(ghs[1].gh_gl, &ul->ul_ut.ut_inum, CREATE, &ip);
1252 if (error)
1253 goto fail_gunlock2;
1254
1255 error = gfs2_inode_refresh(ip);
1256 if (error)
1257 goto fail_iput;
1258
1259 error = gfs2_acl_create(dip, ip);
1260 if (error)
1261 goto fail_iput;
1262
1263 error = link_dinode(dip, name, ip, ul);
1264 if (error)
1265 goto fail_iput;
1266
1267 gfs2_unlinked_put(sdp, ul);
1268
1269 return 0;
1270
1271 fail_iput:
1272 gfs2_inode_put(ip);
1273
1274 fail_gunlock2:
1275 gfs2_glock_dq_uninit(ghs + 1);
1276
1277 fail_gunlock:
1278 gfs2_glock_dq(ghs);
1279
1280 fail:
1281 gfs2_unlinked_put(sdp, ul);
1282
1283 return error;
1284}
1285
1286/**
1287 * gfs2_unlinki - Unlink a file
1288 * @dip: The inode of the directory
1289 * @name: The name of the file to be unlinked
1290 * @ip: The inode of the file to be removed
1291 *
1292 * Assumes Glocks on both dip and ip are held.
1293 *
1294 * Returns: errno
1295 */
1296
1297int gfs2_unlinki(struct gfs2_inode *dip, struct qstr *name,
1298 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1299{
1300 struct gfs2_sbd *sdp = dip->i_sbd;
1301 int error;
1302
1303 error = gfs2_dir_del(dip, name);
1304 if (error)
1305 return error;
1306
1307 error = gfs2_change_nlink(ip, -1);
1308 if (error)
1309 return error;
1310
1311 /* If this inode is being unlinked from the directory structure,
1312 we need to mark that in the log so that it isn't lost during
1313 a crash. */
1314
1315 if (!ip->i_di.di_nlink) {
1316 ul->ul_ut.ut_inum = ip->i_num;
1317 error = gfs2_unlinked_ondisk_add(sdp, ul);
1318 if (!error)
1319 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1320 }
1321
1322 return error;
1323}
1324
1325/**
1326 * gfs2_rmdiri - Remove a directory
1327 * @dip: The parent directory of the directory to be removed
1328 * @name: The name of the directory to be removed
1329 * @ip: The GFS2 inode of the directory to be removed
1330 *
1331 * Assumes Glocks on dip and ip are held
1332 *
1333 * Returns: errno
1334 */
1335
1336int gfs2_rmdiri(struct gfs2_inode *dip, struct qstr *name,
1337 struct gfs2_inode *ip, struct gfs2_unlinked *ul)
1338{
1339 struct gfs2_sbd *sdp = dip->i_sbd;
1340 struct qstr dotname;
1341 int error;
1342
1343 if (ip->i_di.di_entries != 2) {
1344 if (gfs2_consist_inode(ip))
1345 gfs2_dinode_print(&ip->i_di);
1346 return -EIO;
1347 }
1348
1349 error = gfs2_dir_del(dip, name);
1350 if (error)
1351 return error;
1352
1353 error = gfs2_change_nlink(dip, -1);
1354 if (error)
1355 return error;
1356
1357 dotname.len = 1;
1358 dotname.name = ".";
1359 error = gfs2_dir_del(ip, &dotname);
1360 if (error)
1361 return error;
1362
1363 dotname.len = 2;
1364 dotname.name = "..";
1365 error = gfs2_dir_del(ip, &dotname);
1366 if (error)
1367 return error;
1368
1369 error = gfs2_change_nlink(ip, -2);
1370 if (error)
1371 return error;
1372
1373 /* This inode is being unlinked from the directory structure and
1374 we need to mark that in the log so that it isn't lost during
1375 a crash. */
1376
1377 ul->ul_ut.ut_inum = ip->i_num;
1378 error = gfs2_unlinked_ondisk_add(sdp, ul);
1379 if (!error)
1380 set_bit(GLF_STICKY, &ip->i_gl->gl_flags);
1381
1382 return error;
1383}
1384
1385/*
1386 * gfs2_unlink_ok - check to see that a inode is still in a directory
1387 * @dip: the directory
1388 * @name: the name of the file
1389 * @ip: the inode
1390 *
1391 * Assumes that the lock on (at least) @dip is held.
1392 *
1393 * Returns: 0 if the parent/child relationship is correct, errno if it isn't
1394 */
1395
1396int gfs2_unlink_ok(struct gfs2_inode *dip, struct qstr *name,
1397 struct gfs2_inode *ip)
1398{
1399 struct gfs2_inum inum;
1400 unsigned int type;
1401 int error;
1402
1403 if (IS_IMMUTABLE(ip->i_vnode) || IS_APPEND(ip->i_vnode))
1404 return -EPERM;
1405
1406 if ((dip->i_di.di_mode & S_ISVTX) &&
1407 dip->i_di.di_uid != current->fsuid &&
1408 ip->i_di.di_uid != current->fsuid &&
1409 !capable(CAP_FOWNER))
1410 return -EPERM;
1411
1412 if (IS_APPEND(dip->i_vnode))
1413 return -EPERM;
1414
1415 error = gfs2_repermission(dip->i_vnode, MAY_WRITE | MAY_EXEC, NULL);
1416 if (error)
1417 return error;
1418
1419 error = gfs2_dir_search(dip, name, &inum, &type);
1420 if (error)
1421 return error;
1422
1423 if (!gfs2_inum_equal(&inum, &ip->i_num))
1424 return -ENOENT;
1425
1426 if (IF2DT(ip->i_di.di_mode) != type) {
1427 gfs2_consist_inode(dip);
1428 return -EIO;
1429 }
1430
1431 return 0;
1432}
1433
1434/*
1435 * gfs2_ok_to_move - check if it's ok to move a directory to another directory
1436 * @this: move this
1437 * @to: to here
1438 *
1439 * Follow @to back to the root and make sure we don't encounter @this
1440 * Assumes we already hold the rename lock.
1441 *
1442 * Returns: errno
1443 */
1444
1445int gfs2_ok_to_move(struct gfs2_inode *this, struct gfs2_inode *to)
1446{
1447 struct gfs2_sbd *sdp = this->i_sbd;
1448 struct gfs2_inode *tmp;
1449 struct qstr dotdot;
1450 int error = 0;
1451
1452 memset(&dotdot, 0, sizeof(struct qstr));
1453 dotdot.name = "..";
1454 dotdot.len = 2;
1455
1456 gfs2_inode_hold(to);
1457
1458 for (;;) {
1459 if (to == this) {
1460 error = -EINVAL;
1461 break;
1462 }
f42faf4f 1463 if (to == get_v2ip(sdp->sd_root_dir)) {
b3b94faa
DT
1464 error = 0;
1465 break;
1466 }
1467
1468 error = gfs2_lookupi(to, &dotdot, 1, &tmp);
1469 if (error)
1470 break;
1471
1472 gfs2_inode_put(to);
1473 to = tmp;
1474 }
1475
1476 gfs2_inode_put(to);
1477
1478 return error;
1479}
1480
1481/**
1482 * gfs2_readlinki - return the contents of a symlink
1483 * @ip: the symlink's inode
1484 * @buf: a pointer to the buffer to be filled
1485 * @len: a pointer to the length of @buf
1486 *
1487 * If @buf is too small, a piece of memory is kmalloc()ed and needs
1488 * to be freed by the caller.
1489 *
1490 * Returns: errno
1491 */
1492
1493int gfs2_readlinki(struct gfs2_inode *ip, char **buf, unsigned int *len)
1494{
1495 struct gfs2_holder i_gh;
1496 struct buffer_head *dibh;
1497 unsigned int x;
1498 int error;
1499
1500 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh);
1501 error = gfs2_glock_nq_atime(&i_gh);
1502 if (error) {
1503 gfs2_holder_uninit(&i_gh);
1504 return error;
1505 }
1506
1507 if (!ip->i_di.di_size) {
1508 gfs2_consist_inode(ip);
1509 error = -EIO;
1510 goto out;
1511 }
1512
1513 error = gfs2_meta_inode_buffer(ip, &dibh);
1514 if (error)
1515 goto out;
1516
1517 x = ip->i_di.di_size + 1;
1518 if (x > *len) {
1519 *buf = kmalloc(x, GFP_KERNEL);
1520 if (!*buf) {
1521 error = -ENOMEM;
1522 goto out_brelse;
1523 }
1524 }
1525
1526 memcpy(*buf, dibh->b_data + sizeof(struct gfs2_dinode), x);
1527 *len = x;
1528
1529 out_brelse:
1530 brelse(dibh);
1531
1532 out:
1533 gfs2_glock_dq_uninit(&i_gh);
1534
1535 return error;
1536}
1537
1538/**
1539 * gfs2_glock_nq_atime - Acquire a hold on an inode's glock, and
1540 * conditionally update the inode's atime
1541 * @gh: the holder to acquire
1542 *
1543 * Tests atime (access time) for gfs2_read, gfs2_readdir and gfs2_mmap
1544 * Update if the difference between the current time and the inode's current
1545 * atime is greater than an interval specified at mount.
1546 *
1547 * Returns: errno
1548 */
1549
1550int gfs2_glock_nq_atime(struct gfs2_holder *gh)
1551{
1552 struct gfs2_glock *gl = gh->gh_gl;
1553 struct gfs2_sbd *sdp = gl->gl_sbd;
1554 struct gfs2_inode *ip = get_gl2ip(gl);
1555 int64_t curtime, quantum = gfs2_tune_get(sdp, gt_atime_quantum);
1556 unsigned int state;
1557 int flags;
1558 int error;
1559
1560 if (gfs2_assert_warn(sdp, gh->gh_flags & GL_ATIME) ||
1561 gfs2_assert_warn(sdp, !(gh->gh_flags & GL_ASYNC)) ||
1562 gfs2_assert_warn(sdp, gl->gl_ops == &gfs2_inode_glops))
1563 return -EINVAL;
1564
1565 state = gh->gh_state;
1566 flags = gh->gh_flags;
1567
1568 error = gfs2_glock_nq(gh);
1569 if (error)
1570 return error;
1571
1572 if (test_bit(SDF_NOATIME, &sdp->sd_flags) ||
1573 (sdp->sd_vfs->s_flags & MS_RDONLY))
1574 return 0;
1575
1576 curtime = get_seconds();
1577 if (curtime - ip->i_di.di_atime >= quantum) {
1578 gfs2_glock_dq(gh);
1579 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
1580 gh->gh_flags & ~LM_FLAG_ANY,
1581 gh);
1582 error = gfs2_glock_nq(gh);
1583 if (error)
1584 return error;
1585
1586 /* Verify that atime hasn't been updated while we were
1587 trying to get exclusive lock. */
1588
1589 curtime = get_seconds();
1590 if (curtime - ip->i_di.di_atime >= quantum) {
1591 struct buffer_head *dibh;
1592
1593 error = gfs2_trans_begin(sdp, RES_DINODE, 0);
1594 if (error == -EROFS)
1595 return 0;
1596 if (error)
1597 goto fail;
1598
1599 error = gfs2_meta_inode_buffer(ip, &dibh);
1600 if (error)
1601 goto fail_end_trans;
1602
1603 ip->i_di.di_atime = curtime;
1604
d4e9c4c3 1605 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
b3b94faa
DT
1606 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1607 brelse(dibh);
1608
1609 gfs2_trans_end(sdp);
1610 }
1611
1612 /* If someone else has asked for the glock,
1613 unlock and let them have it. Then reacquire
1614 in the original state. */
1615 if (gfs2_glock_is_blocking(gl)) {
1616 gfs2_glock_dq(gh);
1617 gfs2_holder_reinit(state, flags, gh);
1618 return gfs2_glock_nq(gh);
1619 }
1620 }
1621
1622 return 0;
1623
1624 fail_end_trans:
1625 gfs2_trans_end(sdp);
1626
1627 fail:
1628 gfs2_glock_dq(gh);
1629
1630 return error;
1631}
1632
1633/**
1634 * glock_compare_atime - Compare two struct gfs2_glock structures for sort
1635 * @arg_a: the first structure
1636 * @arg_b: the second structure
1637 *
1638 * Returns: 1 if A > B
1639 * -1 if A < B
1640 * 0 if A = B
1641 */
1642
1643static int glock_compare_atime(const void *arg_a, const void *arg_b)
1644{
1645 struct gfs2_holder *gh_a = *(struct gfs2_holder **)arg_a;
1646 struct gfs2_holder *gh_b = *(struct gfs2_holder **)arg_b;
1647 struct lm_lockname *a = &gh_a->gh_gl->gl_name;
1648 struct lm_lockname *b = &gh_b->gh_gl->gl_name;
1649 int ret = 0;
1650
1651 if (a->ln_number > b->ln_number)
1652 ret = 1;
1653 else if (a->ln_number < b->ln_number)
1654 ret = -1;
1655 else {
1656 if (gh_a->gh_state == LM_ST_SHARED &&
1657 gh_b->gh_state == LM_ST_EXCLUSIVE)
1658 ret = 1;
1659 else if (gh_a->gh_state == LM_ST_SHARED &&
1660 (gh_b->gh_flags & GL_ATIME))
1661 ret = 1;
1662 }
1663
1664 return ret;
1665}
1666
1667/**
1668 * gfs2_glock_nq_m_atime - acquire multiple glocks where one may need an
1669 * atime update
1670 * @num_gh: the number of structures
1671 * @ghs: an array of struct gfs2_holder structures
1672 *
1673 * Returns: 0 on success (all glocks acquired),
1674 * errno on failure (no glocks acquired)
1675 */
1676
1677int gfs2_glock_nq_m_atime(unsigned int num_gh, struct gfs2_holder *ghs)
1678{
1679 struct gfs2_holder **p;
1680 unsigned int x;
1681 int error = 0;
1682
1683 if (!num_gh)
1684 return 0;
1685
1686 if (num_gh == 1) {
1687 ghs->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1688 if (ghs->gh_flags & GL_ATIME)
1689 error = gfs2_glock_nq_atime(ghs);
1690 else
1691 error = gfs2_glock_nq(ghs);
1692 return error;
1693 }
1694
1695 p = kcalloc(num_gh, sizeof(struct gfs2_holder *), GFP_KERNEL);
1696 if (!p)
1697 return -ENOMEM;
1698
1699 for (x = 0; x < num_gh; x++)
1700 p[x] = &ghs[x];
1701
1702 sort(p, num_gh, sizeof(struct gfs2_holder *), glock_compare_atime,NULL);
1703
1704 for (x = 0; x < num_gh; x++) {
1705 p[x]->gh_flags &= ~(LM_FLAG_TRY | GL_ASYNC);
1706
1707 if (p[x]->gh_flags & GL_ATIME)
1708 error = gfs2_glock_nq_atime(p[x]);
1709 else
1710 error = gfs2_glock_nq(p[x]);
1711
1712 if (error) {
1713 while (x--)
1714 gfs2_glock_dq(p[x]);
1715 break;
1716 }
1717 }
1718
1719 kfree(p);
1720
1721 return error;
1722}
1723
1724/**
1725 * gfs2_try_toss_vnode - See if we can toss a vnode from memory
1726 * @ip: the inode
1727 *
1728 * Returns: 1 if the vnode was tossed
1729 */
1730
1731void gfs2_try_toss_vnode(struct gfs2_inode *ip)
1732{
1733 struct inode *inode;
1734
1735 inode = gfs2_ip2v_lookup(ip);
1736 if (!inode)
1737 return;
1738
1739 d_prune_aliases(inode);
1740
1741 if (S_ISDIR(ip->i_di.di_mode)) {
1742 struct list_head *head = &inode->i_dentry;
1743 struct dentry *d = NULL;
1744
1745 spin_lock(&dcache_lock);
1746 if (list_empty(head))
1747 spin_unlock(&dcache_lock);
1748 else {
1749 d = list_entry(head->next, struct dentry, d_alias);
1750 dget_locked(d);
1751 spin_unlock(&dcache_lock);
1752
1753 if (have_submounts(d))
1754 dput(d);
1755 else {
1756 shrink_dcache_parent(d);
1757 dput(d);
1758 d_prune_aliases(inode);
1759 }
1760 }
1761 }
1762
1763 inode->i_nlink = 0;
1764 iput(inode);
1765}
1766
1767
1768static int
1769__gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1770{
1771 struct buffer_head *dibh;
1772 int error;
1773
1774 error = gfs2_meta_inode_buffer(ip, &dibh);
1775 if (!error) {
1776 error = inode_setattr(ip->i_vnode, attr);
1777 gfs2_assert_warn(ip->i_sbd, !error);
1778 gfs2_inode_attr_out(ip);
1779
d4e9c4c3 1780 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
b3b94faa
DT
1781 gfs2_dinode_out(&ip->i_di, dibh->b_data);
1782 brelse(dibh);
1783 }
1784 return error;
1785}
1786
1787/**
1788 * gfs2_setattr_simple -
1789 * @ip:
1790 * @attr:
1791 *
1792 * Called with a reference on the vnode.
1793 *
1794 * Returns: errno
1795 */
1796
1797int gfs2_setattr_simple(struct gfs2_inode *ip, struct iattr *attr)
1798{
1799 int error;
1800
1801 if (get_transaction)
1802 return __gfs2_setattr_simple(ip, attr);
1803
1804 error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0);
1805 if (error)
1806 return error;
1807
1808 error = __gfs2_setattr_simple(ip, attr);
1809
1810 gfs2_trans_end(ip->i_sbd);
1811
1812 return error;
1813}
1814
1815int gfs2_repermission(struct inode *inode, int mask, struct nameidata *nd)
1816{
1817 return permission(inode, mask, nd);
1818}
1819